Investing in Crypto Through ETFs and Treasury Companies

On June 27, 2026, CoinDesk published a calculation Strategy Inc. had spent six years making impossible. With MSTR near $82 and bitcoin near $60,000, the company’s enterprise value, its market capitalisation plus its debt and perpetual preferred stock, came to about $50.4B. The 843,000 or so bitcoin it held were worth about $51.1B. For the first time since Michael Saylor bought his first 21,454 coins in August 2020, the market valued the company at less than the coins in it.

Two days later it sold 1,363 bitcoin at an average of $59,256, the start of a 3,588-coin sale that raised about $216M to pay dividends on five series of preferred stock. The premise of the treasury company, that a share of a company holding bitcoin should be worth more than the bitcoin, had failed at the bottom of the cycle, exactly where it mattered most.

Ten weeks later, on September 3, 2026, the spot bitcoin ETFs took in $731M in one session, more than 60% of it into BlackRock’s IBIT, on the same coin, in a wrapper with no debt, no dividends and a 0.25% fee. Between January 2024 and September 2026 the United States built a full set of wrappers around one asset: spot ETFs, futures ETFs, two-times leveraged funds, a converted closed-end trust, a company that is mostly a bitcoin balance sheet, and that company’s own preferred stock. They all track one price. They do not all return it.

This guide is the comparison: what each wrapper charges, tracks, pays in tax and where it breaks, then $25,000 put through the three you are most likely to choose between. Our flagship guide, Investing in Crypto, covers the asset itself, the mining economics, the custody record and the drawdowns, and Investing in Bitcoin and Investing in Ethereum cover each coin on its own. Here we assume you want exposure and ask only how.

There are seven ways to own bitcoin exposure in a US account. Ranked by how many things other than the price of bitcoin can change your outcome, they are the map of this guide, and every later section fills in one rung.

Closest to the coin sits direct ownership: coins bought on an exchange and moved to a wallet you control, with nothing between you and the price but the exchange’s fee and your own competence at holding keys. One rung up is the spot ETF, which puts a sponsor fee, a custodian, an authorised participant and the 9:30-to-4:00 trading day between you and the coin. Above that, the futures ETF, which holds no coins but a ladder of CME contracts, so the shape of the futures curve joins the variables. Above that, the leveraged ETF, which resets daily and so adds the path of the price, not just its end point.

Above that, the closed-end trust, which until January 2024 was the only listed wrapper and traded anywhere from a 40% premium to a discount of nearly 50% to its coins. Above that, the treasury company, a balance sheet of coins financed with converts and preferreds, whose share price is the coin price times a multiple the market sets, and which has ranged from 2.5× to below 1.0× in twenty months. At the top, furthest from the coin, the treasury company’s preferred stock: a fixed claim on a company whose only asset is volatile, paying 8% to 12% for the privilege.

The guide’s first rule follows. Each rung adds a variable you are not paid to hold unless it moves your way, and in 2025 and 2026 most moved the other way. On the same bitcoin over the same period, the difference between the fewest variables and the most has been the difference between losing 20% and losing 75%.

The wrappers have lived through exactly one full cycle, from the ETF launch through the October 2025 peak to September 2026, and what each returned across it is the record every fee and structure argument below has to answer to.

Begin with the coin. Bitcoin was $46,381 when the spot ETFs began trading on January 11, 2024. It reached an intraday high of about $126,200 on October 6, 2025 (SoFi and StealthEX price histories agree on the date). On our tape, which stores one CoinGecko daily close per day and began on August 29, 2025, the highest close was $124,740 on October 7, 2025 and the lowest $58,566 on July 1, 2026, a 53.0% drawdown in under nine months. On September 8, 2026 the tape read $78,576: up 34% from the low, down 37% from the peak, about +69% since the ETF launch, and −30% from the first value we stored.

A spot ETF tracks the coin less its fee and a few basis points of friction. IBIT’s net assets fell from $53.4B on March 31, 2026 to $43.4B on June 30 while the coin fell 13.3%, from $68,130 to $59,101 (both 10-Qs); the rest of the gap was redemptions. That is what tracking looks like: the wrapper moved with the coin and the money moved out.

A futures ETF did worse in both directions. Over the five years to June 2026, bitcoin gained about 85% while ProShares’ BITO lost about 24%, a gap 24/7 Wall St. (June 23, 2026) attributes to the 0.95% fee and to rolling contracts monthly in a market usually in contango. The leveraged funds were worse again. Volatility Shares’ two-times BITX returned −58.3% in its fiscal year to February 28, 2026 (Volatility Shares annual report) and was down 42% for 2026 by late March against 19% for the coin (24/7 Wall St., March 25, 2026). Over the year to January 17, 2026, when MSTR had fallen 53%, the two-times MSTR fund MSTU had fallen 91% (24/7 Wall St.).

The treasury company was worst of all. MSTR peaked at $543 on November 21, 2024 and closed at $137.57 on September 8, 2026, down about 75%, while the coin it holds fell about 20% (Coin Metrics gives $98,546 for the November date). The stock fell almost four times as far as its asset, and Section 9 shows that ratio is the design, not an accident. The wrapper has mattered more than the timing: a buyer at the October 2025 peak holding a spot ETF was down about 37% by September 2026, and the same buyer in MSTR or a leveraged fund was down 60% to 90% on the same bet.

A US spot bitcoin ETF is a grantor trust: it holds bitcoin with a custodian and issues shares that each represent a fixed, slowly shrinking fraction of a coin. The fraction shrinks because the sponsor fee is paid by selling a sliver of the coins each day, which is why, as Section 12 explains, those sales are small taxable events for you. Each piece of the plumbing, the authorised participants, the custodian and the 4:00 p.m. price, is a place where the structure is weaker than it looks.

Shares come into existence through creation and leave through redemption, and only a handful of authorised participants can do either. When investors bid the shares above the value of the coins, an AP delivers coins or cash to the trust, receives new shares and sells them into the demand; when shares trade below, the AP reverses it. That arbitrage pins the share price to the coin.

The SEC’s January 2024 approval allowed only cash creations, so the trust itself had to trade coins and pass the friction to holders. Its order of July 29, 2025 permitted in-kind creations and redemptions, an early policy change under Chair Atkins, and the law firms that summarised it (Dechert, Morrison Foerster, Katten) noted that in-kind is cheaper and more tax-efficient because the trust no longer sells coins to meet a redemption.

Custody is concentrated. Coinbase Custody won eight of the original eleven mandates and, on CryptoSlate’s count for April 8, 2026, was named custodian for funds holding about 84% of US spot bitcoin ETF assets, $77.1B of $91.7B; a stricter count that excludes funds with more than one custodian gives 80.8%, or about $74B. BlackRock added Anchorage Digital as a second IBIT custodian in April 2025, citing the trust’s size, and ARK 21Shares lists Coinbase, BitGo and Anchorage. The coins sit in cold storage in segregated trust accounts, far stronger than an exchange account, but one operational point of failure for four fifths of the category is real.

Price and value can differ. The trust strikes net asset value at 4:00 p.m. Eastern from the CME CF Bitcoin Reference Rate New York Variant, an hour-long average across a set of exchanges, while the shares trade 9:30 to 4:00 on weekdays and the coin trades all weekend, so Monday can gap from Friday’s NAV and fast sessions open a premium or discount. IBIT’s was −0.07% on August 18, 2026 on the sponsor’s figure, and its history shows mostly small premiums, a function of liquidity: IBIT trades tens of millions of shares a day, more than any other fund in the category.

Two features arrived after launch. Options on IBIT began trading November 19, 2024 with a 25,000-contract position limit that the SEC raised to 250,000 in 2025 and, after filings that ran through February 2026, to one million in 2026, the tier that applies to options on Apple and the S&P 500 ETF (Nasdaq ISE’s approval was reported in April, NYSE Arca’s in July). By January 2026 IBIT carried 52% of all bitcoin options open interest, $33B of $65B (CoinDesk, January 13, 2026). And the SEC’s generic listing standards of September 17, 2025 let exchanges list spot funds on other coins without a fund-by-fund order; Bitwise’s Solana fund BSOL began trading October 28, 2025. This guide stays with bitcoin and ether.

Twelve funds hold the same coin, so only two things separate them, fee and liquidity, and the flow record then says who actually owns them.

Twelve US spot bitcoin ETFs traded in September 2026 (Farside’s table lists IBIT, FBTC, BITB, ARKB, BTCO, EZBC, BRRR, HODL, BTCW, MSBT, GBTC and BTC). The sponsor fees, as compiled from the fund pages by U.S. News, CryptoPotato and BitcoinTaxes in 2026, run as follows.

The spread is ten to one for identical exposure: on $25,000, $35 a year at Morgan Stanley’s MSBT against $375 at GBTC, and over ten years about 14% of the average balance at GBTC against 1.5% at the mini trust. Section 7 explains why anyone still holds GBTC. But the cheapest fund is not the most liquid: IBIT at 0.25% is where the market makers, options traders and basis funds live, and for a position you may trade around its depth is worth more than the 10 basis points a year the mini trust saves.

Now the flows. The first year, 2024, brought about $36B of net inflows even as GBTC alone bled more than $21B to cheaper rivals. Cumulative inflows reached $57.7B by late December 2025 on etf.com’s year-in-review, so 2025 added about $21.5B, and here is the number that matters: excluding IBIT, the category saw $3.2B of net outflows in 2025. One fund took all the money. Group assets peaked at $169.48B in October 2025, 6.79% of bitcoin’s market capitalisation, on cumulative inflows that Farside’s data put at about $61.5B and etf.com at $62.77B.

Then the price turned and the money followed it out: about $3.5B in November 2025, $1.1B in December, $1.6B in January 2026, $0.2B in February; $1.3B back in March; $4.1B out in June into the July low, the worst month since launch on CoinDesk’s count of Farside’s tables (some tallies put June at $4.5B). August 2026 brought the turn, about $3.5B in, cutting the year-to-date outflow from $5.3B to $1.8B, and the first week of September added $987M, $731M of it on September 3, about 62% into IBIT. Group assets were $103.34B after that session (HedgeCo, September 4, 2026, from Farside and SoSoValue); cumulative inflows, $54.9B on August 27, were about $6.6B below the peak.

Read those flows against Section 2’s price path and the pattern is plain: the money came in above $100,000 and left below $70,000. The ETF investor buys strength and sells weakness. Part of the inflow was never adoption but the long leg of the basis trade, in which a hedge fund buys the ETF and sells CME futures at a premium that ran 10% to 15% annualised in December 2024 and 4% to 5% by late 2025; that leg unwinds when the premium closes, which is one reason the outflows came in lumps. The wrapper made bitcoin easier to buy and, in equal measure, easier to sell.

The ether wrappers differ from the bitcoin ones on one feature bitcoin lacks: ether pays a yield to holders who stake it, and for eighteen months the ETFs could not collect it.

Spot ether ETFs began trading July 23, 2024, with the bitcoin fee ladder: BlackRock’s ETHA and Fidelity’s FETH at 0.25%, Grayscale’s mini trust ETH at 0.15%, and Grayscale’s converted legacy trust ETHE at 2.50%, the most expensive spot crypto wrapper in the country. None could stake, so a holder paid a fee to own an asset while forgoing its base staking yield of roughly 2.6% to 3% (Investing in Ethereum carries the September 2026 reading), like holding a dividend stock through a broker who kept the dividends.

That changed in three steps. On May 29, 2025 the SEC’s staff said protocol staking is not a securities offering. Grayscale switched staking on in October 2025, and on January 5, 2026 ETHE made the first staking-reward distribution by a US ether exchange-traded product; by its fact sheets of January and April 2026 the mini trust had 67% of assets staked, and the January sheet reported a 4.15% net staking-rewards rate on about $1.79B of assets.

On March 12, 2026 BlackRock launched a separate iShares Staked Ethereum Trust, ETHB, with $107M of seed capital, staking 70% to 95% of its ether through Coinbase Prime and passing about 82% of gross rewards to holders monthly (CoinDesk, March 12, 2026), alongside ETHA, which does not stake.

The products now differ on three terms, and the prospectus is the only place to read them: whether the fund stakes, what share it stakes (unstaked ether meets redemptions), and what cut of the yield the sponsor keeps above the fee. A 4% gross yield with a 0.25% fee and a sponsor share can net a holder 3% or less. Compare funds on the dated net rate, not the fee.

Scale is about a sixth of bitcoin’s. ETHA’s cumulative inflows passed $12B by August 2026, CoinGlass put combined US spot ether ETF net assets at about $15B in early September, and the funds took in $1.42B over nine sessions from August 17, 2026, about $1.02B of it ETHA, with $225.8M on August 27, the largest day since October 28, 2025 (Crypto Briefing; Yahoo Finance). The lesson is the bitcoin lesson: one issuer takes most of the money, and the legacy trust at ten times the fee is the tax-locked exception.

The two wrappers that hold no coins have underperformed the coin in every direction, and the arithmetic shows the loss is structural.

ProShares’ BITO was the first US bitcoin ETF of any kind, launched October 19, 2021, because the SEC would approve exposure through CME futures before it would approve the coin. The fund holds the front-month contract and, as it nears expiry, sells it and buys the next. When the next month costs more than the one expiring, the normal state of the curve, called contango, every roll sells low and buys high, a cost paid whether or not the coin moves.

The fee is 0.95%; the roll cost follows the basis, 10% to 15% annualised at the December 2024 extreme and 4% to 5% by late 2025. 24/7 Wall St. (June 23, 2026) puts BITO’s total drag near 2.9% a year in ordinary conditions, and its five-year result, −24% against the coin’s +85%, is what that drag does over time. There is no reason for a US investor to hold a bitcoin futures ETF outside a covered-call sleeve that specifically wants the futures structure.

The leveraged funds are a different mechanism with a worse outcome. A two-times fund such as Volatility Shares’ BITX or ProShares’ BITU delivers twice the coin’s return each day, and only each day. Across days the reset compounds against you whenever the price moves both ways: bitcoin rises 10% then falls 9.1%, ending flat; the 2× fund rises 20% then falls 18.2%, ending at 98.2, a 1.8% loss on a round trip in which the coin went nowhere. Repeat that a few hundred times a year at 50% to 70% annualised volatility and the decay is the return.

The record bears it out. BITX, whose fiscal-year and 2026 figures are in Section 2, and BITU both lost about 70% over the twelve months to mid-2026 (PortfoliosLab). The single-stock versions, MSTU and MSTX, run at annualised volatilities above 150% (koalagains); MSTU lost 91% in the year to January 17, 2026 while MSTR lost 53%. A leveraged fund on a levered stock on a volatile coin is leverage cubed.

The wrapper that preceded the ETFs failed in the way that matters most here: its price detached from its coins for three years, which is the failure mode that has since caught the treasury companies, and a great deal of money still sits in it.

Grayscale’s Bitcoin Trust began as a private placement in 2013 and traded over the counter from 2015. It was a closed-end fund in all but name: accredited investors could create shares by delivering coins or cash, but nobody could redeem, so the price floated free of the coins. For years it floated above, because it was the only ticker most brokerage and retirement accounts could buy; the premium reached above 40% in early 2021 on CoinGecko’s history, and hedge funds created shares at net asset value, waited out the six-month lock-up and sold at the premium.

Then supply caught up, in February 2021 the price fell below NAV, and it never went back. The discount reached nearly 50% in December 2022 as the firms that had run the arbitrage on borrowed money, Three Arrows Capital and Babel among them, were liquidated after the Terra collapse of May 2022 and their GBTC dumped. An investor who bought at the 2021 premium and sold at the 2022 discount lost the coin’s fall and the whole swing in the premium on top.

The exit came through the courts. On August 29, 2023 the D.C. Circuit ruled the SEC’s refusal to convert the trust arbitrary and capricious; the SEC approved conversion with the other ten funds on January 10, 2024, and around January 26 the discount closed to zero for the first time in nearly three years. Grayscale kept the fee at 1.50%, cut from 2.00%, against 0.19% to 0.39% at most of the newcomers, and the trapped holders finally had a door.

GBTC lost more than $21B in 2024 alone (etf.com), and one fund directory puts cumulative outflows since conversion at about $17.5B and 447,000 coins over a different window; the trust still held about 130,500 coins, roughly $10.1B, on September 1, 2026 (bitcoinetf.directory). Grayscale seeded a mini trust, ticker BTC, at 0.15% with GBTC’s coins so holders could move without a taxable sale.

Three lessons carry forward. A wrapper without redemption can trade at any price, and the premium you pay is not yours to keep. A fee ten times the competition survives only on tax lock-in. And the arbitrageurs who close a premium are the people who, forced to sell, open a discount. All three apply, with larger numbers, to Strategy.

In filings rather than narrative, Strategy Inc. is a count of coins, the paper that bought them, and a bill that must be paid each year whether or not the coins move.

The company formerly called MicroStrategy bought its first 21,454 bitcoin on August 11, 2020 for $250M. By its 8-K for the week to August 30, 2026 it held 845,050 bitcoin acquired for about $63.73B including fees, an average of $75,412 a coin; the last purchase in that filing was 4,603 coins for $369.7M at $80,318, funded by selling common stock at the market. At our tape’s September 8 price of $78,576 the position was worth about $66.4B, 4% above cost, having been about 50% above it a year earlier.

The coins were bought with three kinds of paper. First, common stock sold at the market, the cheapest capital there is while the shares trade above the value of what the proceeds buy. Second, about $8.2B of convertible notes on CoinDesk’s January 22, 2026 tally, at a weighted coupon of 0.421%, close to free while the stock rises through the conversion prices, and plain debt to refinance if it does not. Third, from January 2025, five series of perpetual preferred stock that became the main funding source and the main cost.

STRK pays 8.00% and converts into common; STRF pays 10.00% and ranks senior; STRD pays 10.00% without accruing missed dividends; STRE is the euro version at 10.00%; and STRC, the “Stretch” preferred, pays a variable rate the company sets to hold the shares near $100 par, and has only ever been raised or held: from 9.00% at its July 2025 launch through seven monthly increases to 11.50% by March 2026, held there through June (8-K of May 30), and 12.00% from July 1 (8-K of June 29), held for August (CoinDesk, April 1, 2026; company 8-Ks). Shareholders approved semi-monthly STRC payments on June 8, 2026.

The stack grew faster than the coins. CoinDesk’s January 2026 tally put preferred notional at $8.36B, with STRC at $3.4B; by the June 30, 2026 10-Q STRC’s notional was $10.49B and STRF’s $1.28B. On our arithmetic STRC alone at 12.00% costs about $1.26B a year, and with STRK, STRD and STRF at their disclosed notionals the four dollar series cost roughly $1.6B a year before STRE, whose €775M at 10.00% (10-Q for March 31, 2026) adds about $90M: about 2.5% of the coins, every year, in cash the coins do not generate.

The company’s answer has been reserves: a USD Reserve of $5.10B and a USD Cash pool of $1.44B on September 7, 2026, built partly from $2B of common sold in the week to August 23 with no coins bought (The Block, August 24, 2026), and, in June and July, from selling coins.

Think of the preferred stack as the company’s expense ratio: IBIT charges 0.25% of the coins a year, and Strategy must find about 2.5%, plus the converts’ coupon and the cost of a listed company, by issuing stock, drawing reserves or selling coins. Which of those it can afford depends on one multiple, and the next section turns it into a number.

One number decides whether a treasury stock is worth more or less than its coins. Here is how to compute it from filings, and what it did to Strategy’s holders from November 2024.

mNAV, the multiple of net asset value, is the market’s price for a company’s coins divided by the coins’ own price. The simple version divides market capitalisation by the value of the bitcoin held. The version that matters, enterprise mNAV, uses enterprise value, market capitalisation plus debt plus preferred at notional, less cash, because those holders have a claim on the coins ahead of the common. At 2.0× a company sells a dollar of bitcoin for two dollars; at 0.9× it is a dollar of bitcoin for ninety cents, less whatever it costs to keep the dividends paid.

The premium is the fuel. At 2.0×, selling $1B of stock buys $1B of coins the market values at $2B, so every issuance raises the bitcoin behind each existing share, which the company reports as “BTC Yield” and which is real while the multiple holds. At 1.0× issuance is neutral. Below 1.0× every share sold to pay a dividend hands away more bitcoin than the cash it raises, so the rational course is to stop issuing and buy back paper. The model runs on the premium and stalls without it.

The history: above 2.5× in November 2024, the week the stock peaked at $543; about 1.0× to 1.2× by November 2025; and on June 26 and 27, 2026, the reading in the cold open, $50.4B of enterprise value against $51.1B of coins, or 0.99×, the first below one (The Block; CoinDesk). The equity-only version stayed higher through the summer because it ignores some $23B of debt and preferreds, which is why you should distrust any mNAV figure that does not say which one it is.

Now the leverage, on our arithmetic from the latest filings, rounded, ignoring STRE, the software business and operating costs. Coins: 845,050 at $78,576, about $66.4B. Converts: $8.2B. Preferred at notional: $14.6B. Cash and reserves: $6.5B. At exactly 1.0× enterprise mNAV the common is worth 66.4 − 8.2 − 14.6 + 6.5, about $50.2B, so each dollar of equity carries $1.32 of bitcoin. Add the dividend bill of about $1.64B a year, paid from that equity in one form or another, and run three outcomes over three years with the multiple held at 1.0×:

• Bitcoin +50%: coins $99.6B, equity $78.5B after $4.9B of dividends, up 56%. The leverage beats the carry.
• Bitcoin flat: equity $45.3B, down 10%. The carry alone costs a tenth of the equity in three years.
• Bitcoin −50%: coins $33.2B, equity $12.1B, down 76%. The claims ahead of the common do not shrink when the coins do.

That is the shape of MSTR against bitcoin on this cycle: a 75% fall against a 20% fall, which our model attributes to three things stacked, the multiple going from 2.5× to 1.0× (a 60% loss on its own), the fixed claims levering the rest, and the carry. All three were visible in the filings before they happened.

Once the premium was gone, Strategy’s own 8-Ks record what it did, in order, and the sequence is the template for any treasury company when the fuel runs out.

The first sale was small and symbolic. Between May 26 and 31, 2026 the company sold 32 bitcoin for about $2.5M at an average of $77,135, its first sale since December 2022, to fund preferred dividends; the filing said it would manage its balance sheet actively rather than never sell, MSTR fell 5.85% and bitcoin fell 2% to its lowest since mid-April (CNBC, June 1, 2026; Arkham). STRC, meant to trade at $100, hit a record low of $89 on June 18 (CoinDesk), the market saying 11.50% was not enough; the rate went to 12.00% from July.

The second sale was the largest in the company’s history: from June 29 to July 5, 3,588 bitcoin for about $216M, including 1,363 coins at $59,256 on June 29 and 30, within a few dollars of the cycle low, to cover the second-quarter dividends on STRF, STRE, STRK and STRD and the June STRC payment, leaving 843,775 coins and $2.55B of USD Reserves on July 5 (8-K; The Defiant). The two sales together, 3,620 coins, are under half a percent of the holdings, which is the point: the company did not need to raise much, it sold because issuing stock below the value of the coins was worse.

Then the pivot. In the week to August 23 it sold about $2B of MSTR, bought no bitcoin and set up a $1.6B “USD Cash” pool (The Block, August 24); the next week it bought 4,603 coins; and in the week to September 7 it bought none, sold no stock, and repurchased 1,810,885 STRC shares for $176.3M under an authorisation the board had doubled to $2B, leaving $1.19B available, with $1.0B more authorised for MSTR, a $5.10B USD Reserve and $1.44B of USD Cash (8-K of September 8, 2026). The stock fell on the day.

Read it as a machine reversing. In 2024 the company sold stock at a premium to buy coins; in 2026 it sold coins to pay the holders of the paper that bought the coins, sold stock to build a cushion, then used the cushion to buy back the paper. The company was, by September, more liquid than in June, but bitcoin per share, the number the model was built to raise, is flat at best, and the common holder pays about 2.5% of the coins a year to watch it.

The one external threat, MSCI’s October 2025 consultation on ejecting companies with more than half their assets in digital assets from its indexes, went the company’s way on January 6, 2026, when MSCI deferred any exclusion pending a broader consultation and the stock rose 6% after hours; the $10B to $15B of estimated index selling did not happen. It remains on the list of things that could.

A hundred-odd companies copied Strategy in 2025, and their unwind, with cases named and dated, is the clearest evidence of what the model does at the bottom of a cycle.

The wave was 2025: Metaplanet in Tokyo, Twenty One Capital with Tether, SoftBank and Cantor Fitzgerald behind it (about 43,500 coins on The Block’s August 2026 tally), Trump Media, GameStop, Semler Scientific, a dozen shells reverse-merged into “digital asset treasuries”, and an ether version led by BitMine Immersion and SharpLink, all on the same premise of selling stock above the coins to buy more coins. DL News put Metaplanet’s premium at 237% in July 2025; by late October 2025 the multiple was 0.88×, a 12% discount that prompted the buyback below (CoinDesk, October 28, 2025), and by June 2026 the stock was down 42% for the year and 85% over twelve months (CoinMarketCap Academy).

The unwind ran through 2026, and CoinDesk’s July 24 survey names the sellers. Sequans, a French chipmaker that turned itself into a bitcoin treasury in July 2025, sold 970 coins in November 2025 to redeem half its convertible debt, cutting it from $189M to $94.5M, sold 1,025 more in the first quarter of 2026, and by May 2026 was down to 658 coins from about 3,100 before the first sale, declaring the treasury strategy over (Sequans; CoinDesk, November 4, 2025; The Block, May 28, 2026).

Nakamoto Holdings sold 284 coins for $20M in March 2026 at about $70,422, 40% below its $118,171 average cost, to fund operations, and then reported a $238.8M net loss for the first quarter of 2026, $102.5M of it a mark-to-market loss on the 5,058 coins it kept (CoinDesk, March 31, 2026; Nakamoto Q1 results, May 13, 2026).

Satsuma Technology’s shareholders approved liquidating all 668 coins and delisting from London. Smarter Web Company sold 178 coins to repay a convertible; Empery Digital sold about half its stack to repay debt and fund buybacks (CoinDesk, July 11, 2026); K Wave Media sold its 88 coins to repay $6M of notes and exited; Strategy, as Section 10 records, sold 3,620.

The mechanism is reflexive in both directions. A paper from the top of the wave, “How Digital Asset Treasury Companies Can Survive Bear Markets” (arXiv 2511.01135, November 2025), formalises a “no-forced-sale” test: whether a company’s cash flows can bridge an 18-to-24-month bear market without selling coins. The 2026 record is what failing that test looks like. A premium lets a company buy accretively; a discount removes the reason to issue; fixed claims force sales; sales at the low lock in the loss; the loss deepens the discount.

Metaplanet’s answer is to borrow against its coins rather than sell them, with a buyback of up to 150 million shares, 13.13% of the company, financed by a bitcoin-collateralised facility of up to $500M and running to October 28, 2026 (Cointelegraph; The Block). That converts an equity discount into balance-sheet leverage, a different risk, not a smaller one. Its holdings passed 43,000 coins on July 2, 2026 (CoinDesk), third after Strategy and Twenty One, and its multiple depends on who computes it: 0.86× on the enterprise read CoinDesk published on September 4, 2026, 1.21× on mnav.com’s equity-based live figure on September 9. The gap is the debt.

The ether version has not unwound, and that should be said plainly. BitMine bought 28,086 ether in the week to September 7, 2026, with ether below $2,500, taking its holdings to 5.93 million ether, about 4.9% of supply and $14.8B, at an average cost of $2,495 across the position; it has bought every week since June 30, 2025 (The Block and CoinDesk, September 8, 2026; 24/7 Wall St., September 9).

Staked ether pays a yield, so an ether treasury can cover part of its cost from the asset rather than from issuance, a real structural difference. What it cannot escape is the multiple: a treasury company at 1.0× is a closed-end fund with a management team, and at 0.8× it is one with a discount.

The same bitcoin outcome leaves three different after-tax sums depending on the paper you held it in. This is the treatment in a US account as of September 2026.

The rates are common to all three. Gains on positions held more than a year are taxed at long-term rates, for 2026 (IRS Revenue Procedure 2025-32, October 9, 2025) 0% up to $49,450 of taxable income single and $98,900 joint, 15% up to $545,500 and $613,700, and 20% above; the 3.8% net investment income tax of section 1411 applies above $200,000 of modified adjusted gross income single and $250,000 joint, so the top effective rate is 23.8%. Held a year or less, gains are ordinary income up to 37%.

Bitcoin is not a collectible under section 408(m), so the 28% rate on the art, wine and cars elsewhere in this series does not apply, and neither does the bar on collectibles in an IRA.

The spot ETF

The trust is a grantor trust, so the law looks through it: you own your share of the coins, with the same holding period and rates as if you held them yourself. Each year the trust sells slivers of coin to pay its fee, and your share of those sales is a small gain or loss reported in a year-end tax information statement even if you sold nothing. Your broker reports your own sales on Form 1099-B, because the 1099-DA regime excludes most grantor-trust crypto ETFs (The Tax Adviser, March 2026).

The open point is the wash-sale rule of section 1091, which applies to “securities”: brokers treat ETF shares as securities and adjust losses for repurchases within thirty days, while a look-through reading says the shares are bitcoin and exempt; Green Trader Tax notes the IRS has not ruled. Assume it applies, because your broker will, and harvest by switching to another issuer’s fund.

The treasury stock

MSTR is a stock and nothing about bitcoin’s treatment reaches it: capital gains at the rates above, the wash-sale rule without ambiguity, no fee-sale events. The preferreds are stock too; their dividends arrive on Form 1099-DIV and their character, ordinary, qualified or return of capital, depends on the company’s earnings and profits, reported after year end on Form 8937. Do not assume the 12% is qualified income until you have seen that form.

The coin

Bitcoin held directly is property under IRS Notice 2014-21: capital gains at the same rates, and no wash-sale rule, because section 1091 reaches securities and not property, so you may sell at a loss and rebuy the same minute. As of September 9, 2026 none of the bills to close that gap, Senator Lummis’s digital-asset tax bill, the PARITY Act or Representative Arrington’s June 2026 House bill, had passed, though Treasury’s fiscal 2025 Greenbook scored closing it at about $23.5B over a decade and it eventually will.

Reporting changed: exchanges issued Form 1099-DA with gross proceeds for 2025 sales and must report cost basis from the 2026 tax year, but only for covered assets, coins acquired on or after January 1, 2026 and held in the same broker’s account until sold; anything bought earlier, transferred in or sold on a decentralised exchange is reported without basis, and the IRS presumes zero unless you prove otherwise. Basis is tracked wallet by wallet under Revenue Procedure 2024-28 from January 1, 2025. The freedom to harvest is worth money in a drawdown; the record-keeping is its price.

The account

The cleanest structure in this guide is a spot ETF inside a traditional or Roth IRA, where none of the above applies until withdrawal, or ever in the Roth; every major brokerage allows it. Self-directed “crypto IRAs” that hold coins charge for it: iTrustCapital 1% a trade, Alto 1% a trade, BitcoinIRA up to 5.99% on the first purchase plus an annual fee of up to 2% (IRA Financial and LendEDU, 2026), rarely worth it against a 0.15% to 0.25% ETF.

Put the same $25,000 into IBIT, MSTR and coins held directly on September 8, 2026, hold three years, and run three bitcoin outcomes; what follows is what each wrapper leaves in your pocket.

The assumptions. Bitcoin is $78,576 at entry; MSTR is $137.57 and trades at 1.0× enterprise mNAV throughout, which Section 9 shows is generous; the investor pays 15% federal and 5% state on long-term gains, 20% together, below the NIIT threshold. The outcomes at the end of year three are bitcoin at $117,864 (+50%), $78,576 (flat) and $39,288 (−50%). IBIT charges 0.25% on the average balance plus $5 of spread each way. MSTR uses Section 9’s model. Figures are rounded.

The direct route pays Coinbase Advanced’s 0.60% maker rate on a limit order each way, $5 to withdraw and $150 for a hardware wallet and steel backup. The published schedule’s lowest tier is 1.20% taker and 0.60% maker under $1,000 of trailing volume, where a first purchase lands; several 2026 guides still quote an older 0.60% and 0.40% entry tier, so check the page the day you trade.

IBIT in a taxable account

Up 50%: shares worth $37,500 less about $234 of fees and $10 of spread, $37,256; gain $12,256, tax $2,451, net $34,805. Flat: $24,802 after $188 of fees, a $198 loss, no tax. Down 50%: $12,349, with a $12,651 realised loss deductible against other gains, worth about $2,530 at 20%, subject to the thirty-day rule. In a Roth IRA all three are kept whole.

Coins, exchange to hardware wallet

The $150 buy fee leaves $24,850 of coin. Up 50%: coins worth $37,275, sale fee $224, proceeds $37,051; basis is the full $25,000 because the buy fee capitalises, gain $12,051, tax $2,410, and after the $155 of device and withdrawal the net is $34,486, about $320 behind IBIT. Flat: $24,546. Down 50%: $12,195, but the $12,650 loss can be harvested by selling and rebuying within the hour with no wash-sale wait, the one place the coin beats the ETF.

Fees were $529 against IBIT’s $244, a small price for sovereignty. The larger differences are the ones no arithmetic captures: the coin holder can be phished, robbed or die with the seed, and must keep every transfer record for the 1099-DA reconciliation.

MSTR

$25,000 at 1.0× enterprise mNAV buys equity carrying about $33,000 of bitcoin, $8,000 of fixed claims and a dividend bill. Up 50%: equity up about 56% to $39,000, gain $14,000, tax $2,800, net $36,200, about $1,400 ahead of IBIT, which is what leverage is for. Flat: the carry alone takes the equity down about 10% to $22,550; if the multiple slips to 0.8× on a flat coin, about $16,000, because the fixed claims come off a smaller enterprise value. Down 50%: the fixed claims do not shrink and the equity falls about 76% to $6,000, against $12,349 in the ETF, and that is with the multiple held at 1.0×, which in the 2026 drawdown it was not.

Side by side, the ETF and the coin finish within $320 of each other in every outcome; the choice between them is about custody, harvesting and records, not money. MSTR is a different instrument: about $1,400 better than the ETF if the coin rises by half, about $6,300 worse if it halves, with a multiple that has not, in practice, held constant. It is not a way to own bitcoin. It is a way to borrow against bitcoin at 8% to 12% through someone else’s balance sheet: a levered bet with a carry.

For an outsider putting money into these wrappers for the first time, this is the sequence, in the order that keeps the mistakes cheap.

1. Decide the account before the wrapper. If you have unused Roth or traditional IRA room, the spot ETF goes there first; the fee-sale events, the wash-sale question and the capital gains all disappear.
2. Pick the fund on fee and scale, once. Grayscale’s mini trust at 0.15% or Morgan Stanley’s MSBT at 0.14% for a position you will not trade; IBIT at 0.25% for one you will, or for options. Never GBTC or ETHE. Check the fee on the issuer’s page the day you buy; VanEck’s waiver ended July 31, 2026 and others will change.
3. Trade during US hours with a limit order, after 10:00 a.m. Eastern and never in the first half hour after a weekend move, when the shares can gap from the 4:00 p.m. NAV.
4. If you want ether, read the staking terms. The net rate, the share staked and the sponsor’s cut are in the prospectus and fact sheet; ETHB and Grayscale’s mini trust pass yield through, ETHA does not.
5. Skip the futures and leveraged funds unless you can name the day you will sell.
6. If you want coins directly, buy on Coinbase Advanced or Kraken Pro with a limit order, withdraw to a hardware wallet the same day, record the transfer, and set up a tax package (Koinly, CoinTracker or CoinLedger) before the first trade, because the 1099-DA regime will surface every gap from the 2026 tax year.
7. If you want a treasury company, compute its enterprise mNAV from the latest 10-Q and write the number down; set a sale rule at the multiple, not the price; read the dividend schedule and the cash position; and size for a 75% drawdown, which is what it did last time.
8. If you want the preferreds, know that STRC’s rate is set by the company to hold $100 and has only ever been raised or held, that STRD does not accrue missed dividends, and that all of them are claims on a company whose only asset fell 53% in nine months. They are high-yield credit on volatile collateral, priced as such.
9. Date-stamp everything. The fees, flows, multiples and thresholds here are as of September 9, 2026, and all of them move.

These are the readings that would change the view in this guide, with the thresholds at which they matter.

• Strategy’s enterprise mNAV against 1.0×. Above 1.2× the company resumes accretive issuance and the treasury trade works again; a second break below 1.0× with STRC at 12% or above means more coin sales. Compute it weekly from the Monday 8-K.
• STRC against $100. The June 18, 2026 low of $89 preceded the largest coin sale in the company’s history by eleven days. A trade below $95 with the rate already at 12% is the leading indicator; a rate cut would be the first in the series’ history.
• The USD Reserve and Cash pool. $5.10B and $1.44B on September 7, 2026 cover about four years of the dollar preferred dividends on our arithmetic. Below two years of cover, coin sales return.
• Monthly ETF flows against the price. Three consecutive months of inflows with the coin below $80,000 would be the first evidence since launch of ETF money buying weakness; $3B or more out in a month, the November 2025 pattern, says the marginal holder still follows the trend.
• The ex-IBIT flow. The category outside BlackRock lost $3.2B in 2025. Two consecutive positive quarters would mean the ownership base has broadened; until then, one fund’s clients are the market.
• Custody concentration. Coinbase held about 84% of ETF assets in April 2026. A second large issuer going multi-custodian, or a regulatory finding against any custodian, changes the risk file for every fund at once.
• The wash-sale bills. Passage of the Lummis, PARITY or Arrington bill removes the one tax advantage of direct ownership over the ETF and makes IBIT in a Roth the nearly universal answer.
• The ether treasuries’ coverage. BitMine has bought ether every week for sixty-two weeks; the first week it does not is the news.
• Our tape. Bitcoin closed at $78,576 on September 8, 2026 and the Crypto sub-index at 59.052, up 32.4% in thirty days and down 41.5% over a year. A close above the $124,740 peak of October 7, 2025 puts every ETF buyer of the cycle above water; a close below $58,566 puts the treasury companies back where they were in July.